[
https://issues.apache.org/jira/browse/HBASE-10803?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13944318#comment-13944318
]
LiJianyin commented on HBASE-10803:
-----------------------------------
I used our data from online(asyn) to test. The mothods just as bellow, maybe
you can try:
public boolean addOrUpdate(String rowkey, Long mid, String value, long
time){
HTableInterface table = null;
try{
table = connection.getTable(TABLE_NAME);
Put p = new Put(Bytes.toBytes(rowkey));
p.add(FAMILY, Bytes.toBytes(mid), Bytes.toBytes(value));
table.put(p);
table.close();
return true;
} catch (Exception e){
logger.warn("hbase put failed!",e);
return false;
}
}
when the method "addOrUpdate" returns true, write the params (rowkey, mid,
String , time) to PUT_SUCCESS_LOG; when it returns false, ignore it.
The addOrUpdate method about 200-500 times/s
//delete and put atomically on a single row
public boolean delete(String rowkey, Long mid, String value, long time){
try{
HTableInterface table = connection.getTable(TABLE_NAME);
Delete d = new Delete(Bytes.toBytes(rowkey), time);
d.deleteColumns(FAMILY, Bytes.toBytes(mid), time);
Put p = new Put(Bytes.toBytes(rowkey));
p.add(FAMILY_DEL, Bytes.toBytes(mid), time,
Bytes.toBytes(value));
RowMutations rm = new
RowMutations(Bytes.toBytes(rowkey));
rm.add(d);
rm.add(p);
table.mutateRow(rm);
table.close();
return true;
} catch (Exception e){
logger.warn("hbase delete failed!",e);
return false;
}
}
when the method "delete" returns true, write the params (rowkey, mid, String ,
time) to DEL_SUCCESS_LOG; when it returns false, ignore it.
The delete method about 50-120 times/s
Both addOrUpdate and delete were running for 15+ days, the two types of method
maybe operate the same rowkey and mid.
Then check every record in PUT_SUCCESS_LOG and DEL_SUCCESS_LOG whether the
method success. PUT_SUCCESS_LOG all right, but few of DEL_SUCCESS_LOG wrong,
and the wrong time just restarting or balancing.
Thanks.
> "mutateRow" maybe fail when Region Server restarted(or killed) or Region
> Balance
> --------------------------------------------------------------------------------
>
> Key: HBASE-10803
> URL: https://issues.apache.org/jira/browse/HBASE-10803
> Project: HBase
> Issue Type: Bug
> Affects Versions: 0.96.1.1
> Environment: hbase0.96.1.1
> hadoop2.2
> jdk1.7
> 5 Region Server
> Reporter: LiJianyin
>
> found that the operation "mutateRow" maybe fail if Region Server
> restarted(or killed) or Region Balance, but the java client didnot throw
> exceptions and there was no exception in the server's log either.
> the client code :
> HTableInterface table = connection.getTable(TABLE_NAME);
> Delete d = new Delete(Bytes.toBytes(rowkey), time);
> d.deleteColumns(FAMILY, Bytes.toBytes(id), time);
> Put p = new Put(Bytes.toBytes(rowkey));
> p.add(FAMILY_DEL, Bytes.toBytes(id), time,
> Bytes.toBytes(value));
> RowMutations rm = new
> RowMutations(Bytes.toBytes(rowkey));
> rm.add(d);
> rm.add(p);
> table.mutateRow(rm);
> table.close();
> the test data:
> Operator Type: put; mutateRow(delete+put)
> Operator Total Count:202194822 put:160866148 mutateRow:41328674
> All put correct, but 161 mutateRow incorrect (161 data was not
> mutateRowed in fact).
> then we checked log when the time mutateRow failed, NO EXCEPTION
> THROWED, found that as follow:
>
> 39 failed hbase auto region-balance
> 44 failed hbase region server killed
> 59 failed hbase region server restart
> 19 failed hadoop datanode restart
--
This message was sent by Atlassian JIRA
(v6.2#6252)
